In composite video television systems such as NTSC and PAL, luminance and chrominance information share a portion of the total signal bandwidth. While clean separation between luminance and chrominance is highly desired, current widely used video signal decoders can misinterpret the residues of shared luminance and chrominance after separation, resulting in cross color and dot crawl. Both are highly disturbing artifacts. The term “cross color” refers to corruption of the chrominance spectrum caused by the misinterpretation of high-frequency luminance information as chrominance information. Cross color manifests itself in spectrum of bright colors changing from frame to frame. Conversely, the term “dot crawl” or “cross luminance” refers to corruption of the luminance spectrum by the misinterpretation of chrominance information as high-frequency luminance information. Dot crawl manifests itself in patterned high amplitude noise.
Both artifacts can be reduced by selectively filtering video signals during signal processing. The filtering process usually employs a 3D comb filter comprising at least one line comb filter and at least one frame comb filter. A line comb filter can reduce such artifacts but its effectiveness is limited to artifacts generated by vertical edges and it has a disadvantage of decreasing the vertical resolution. A frame comb filter, on the other hand, provides maximum picture resolution but can only be applied to stationary parts of a picture. To maximize the effectiveness of the comb filters, a highly precise motion detector that can differentiate between the moving and stationary pixels is required.
Conventional arts use a low pass inter-frame difference to generate a motion map to select line comb filters when motion is detected and frame comb filters when there is no motion. Depending on the cut-off frequency of the low pass filter, the performance of the 3D comb filter varies. If the cut-off frequency is high, some motion due to cross luminance may be falsely detected and the 3D comb filter's effectiveness is reduced. If the cut-off frequency is low, motion with higher frequency content may not be detected and motion smearing results. The higher the overlapping of the chrominance with video bandwidth, the more ineffective the motion detection.
Some have improved the performance of motion detection by associating oblique correlation with likelihood of false motion. One disclosed motion detection device including an oblique correlation detection section, motion detection section and motion determination section decreases the sensitivity of motion detection in the presence of an oblique correlation. However, the implementation of the concept using decreased sensitivity in presence of oblique correlation is not sufficient because of the conflict of interests. On one hand, the decreased sensitivity may have impaired the detection of true motion for oblique objects. On the other hand, decreased sensitivity may not be sufficient to prevent false motion detection in mixed color/luminance edges since cross luminance are typically of large amplitudes.
Another example for motion detection uses a plurality of temporal pixels to determine the motion or still status of the video composite signal suitable for use in a 3D comb filter in video decoder. Yet another example for motion detection uses a motion detection circuitry with precise Y motion and C motion detection. The Y motion detection uses the frame difference of line-comb Y signal with chroma level and vertical edge consideration. The C motion detection uses the frame difference of line-comb C signal, together with the frame difference of composite signal and chroma vertical and horizontal correlation computed from the frame-comb Y signals of adjacent lines. Yet another example for motion detection uses a two-frame difference signal that has been filtered to remove chrominance information. The filtering is performed on at least one spatial axis according to the spatial correlation.
Generally, these methods do not consider motion contributed by chroma component because of interfering high frequency luminance at chroma band. However, there exists motion with purely color motion and a misdetection results in color smearing.
Still some use the two-frame difference to detect motion. Although this motion detection considers the contributions from both luminance and chroma, it does not represent the temporal difference between the frames being filtered and may not be sufficient for the PAL standard.
FIG. 1 shows a typical prior art motion detector 700 that is adapted for the PAL standard. The pixel difference Pn,n−2 between frame n and frame n−2, and the pixel difference Pn,n−4 between frame n and n−4 are computed by subtractors 701 and 702 respectively. Due to the inverted phase difference between frame n and n−2, a selection by a selector 707 of a vertical low pass filter 705, a horizontal low pass filter 704 and a vertical horizontal low pass filter 703 is performed by a correlator 706 to make the pixel values independent of cross colour. The final motion value is the maximum absolute pixel difference Pn,n−2 and absolute filtered pixel difference Pn,n−4.